Integrating eQTL data with GWAS summary statistics in pathway‐based analysis with application to schizophrenia

Abstract

Many genetic variants affect complex traits through gene expression, which can be exploited to boost statistical power and enhance interpretation in genome-wide association studies (GWASs) as demonstrated by the transcriptome-wide association study (TWAS) approach. Furthermore, due to polygenic inheritance, a complex trait is often affected by multiple genes with similar functions as annotated in gene pathways. Here, we extend TWAS from gene-based analysis to pathway-based analysis: we integrate public pathway collections, expression quantitative trait locus (eQTL) data and GWAS summary association statistics (or GWAS individual-level data) to identify gene pathways associated with complex traits. The basic idea is to weight the SNPs of the genes in a pathway based on their estimated cis-effects on gene expression, then adaptively test for association of the pathway with a GWAS trait by effectively aggregating possibly weak association signals across the genes in the pathway. The P values can be calculated analytically and thus fast. We applied our proposed test with the KEGG and GO pathways to two schizophrenia (SCZ) GWAS summary association data sets, denoted by SCZ1 and SCZ2 with about 20,000 and 150,000 subjects, respectively. Most of the significant pathways identified by analyzing the SCZ1 data were reproduced by the SCZ2 data. Importantly, we identified 15 novel pathways associated with SCZ, such as GABA receptor complex (GO:1902710), which could not be uncovered by the standard single SNP-based analysis or gene-based TWAS. The newly identified pathways may help us gain insights into the biological mechanism underlying SCZ. Our results showcase the power of incorporating gene expression information and gene functional annotations into pathway-based association testing for GWAS.

Integration of Enhancer-Promoter Interactions with GWAS Summary Results Identifies Novel Schizophrenia-Associated Genes and Pathways

“…We used a stringent Bonferroni cutoff (0:05=500 ¼ 1 3 10 24 ) for pathway-based analysis. For comparison, we applied a new method (Wu and Pan 2018), which extends TWAS from gene-based to pathwaybased analysis. Briefly, we applied the weighted SPU(1) and SPU(2) tests, in which each of the SNPs in the genes (or their extended regions) belonging to a pathway is weighted by its estimated cis-effect size on the gene expression based on an eQTL data set.…”

An Optimally Weighted Combination Method to Detect Novel Disease Associated Genes Using Publicly Available GWAS Summary Data

“…If we focus on data from transcriptomewide association studies, we can set the weights of WSS and the elements of the diagonal of matrix A as the corresponding estimated cis-effects on gene expression. Then, the two methods WSS and SSU contained in our method become PathSPU(1) and PathSPU(2) method proposed by Wu and Pan (2018). Therefore, the OWC is an optimally weighted combination test that can reach maximized power.…”

“…If we believe that the causal SNPs would be subject to "purifying selection" and thus appear less frequently in the population than neutral SNPs, we set w m = 1/ p m (1 − p m ), where p m denotes the minor allele frequency (MAF) of the m th variant, and obtain L W = L(1/ p 1 (1 − p 1 ), · · · , 1/ p M (1 − p M )), which is the weighted sum statistic (WSS) (Madsen and Browning, 2009). If we assume that the values of the weights W come from gene expression or functional annotation data, the test degenerates into the PathSPU(1) test (Wu and Pan 2018). We know that S(w 1 , · · · , w M ) follows central chi-square distribution with 1 degree of freedom (χ 2 1 ) and L(w 1 , · · · , w M ) follows multivariate normal distribution with mean 0 and covariance matrix W RW under the null hypothesis when the choice of the weight function W is not proportional to Z.…”

“…Based on the asymptotic null distribution of Z in Equation (2), the test L Q = Z Z follows a mixture of chi square distribution under the null hypothesis: L Q ∼ M m=1 λ m χ 2 1 , where λ 1 , · · · , λ M are the eigenvalues of R. Particularly, if we set the diagonal element of A as the beta distribution density function with pre-specified shape parameters as 1 and 25 evaluated at the corresponding sample MAF in the data, it degenerates into the sequence kernel association test (SKAT) for rare variants (Wu et al, 2011). If the value of the diagonal elements of A comes from a set of gene expression derived weights, it degenerates into PathSPU(2) test method (Wu and Pan, 2018). Naturally, these two methods (SKAT and…”

Incorporating external information to improve sparse signal detection in rare‐variant gene‐set‐based analyses

“…A limitation of all of the above approaches is that they treat all genes within the focal gene‐set equally, even though there is substantial external prior information about the relative importance of a gene within a gene‐set and incorporating such prior information may substantially increase power (Genovese, Roeder, & Wasserman, 2006; Ma & Wei, 2019; Pan, Kwak, & Wei, 2015; Roeder & Wasserman, 2009; Wu & Pan, 2018) Genic intolerance (Petrovski, Wang, Heinzen, Allen, & Goldstein, 2013), network centrality (Barabási & Albert, 1999; Goh et al, 2007; White & Smyth, 2003), and gene expression in disease relevant tissues are examples of sources of prior information that can be used to quantify the relative importance of genes within gene‐sets. We discuss these sources of prior information in detail in the materials and methods section below.…”